arXiv:2608. 06776v1 Announce Type: new Abstract: A standard self-attention layer consists of two interacting circuits: the query-key circuit that governs attention allocation, and the output-value circuit that maps attended representations to predictions.
arXiv:2605. 09778v2 Announce Type: replace Abstract: Evaluating softmax attention over a fixed long context requires reading every cached key-value pair for each new query token.
arXiv:2606. 01294v1 Announce Type: cross Abstract: Linear attention reduces the quadratic cost of softmax attention by maintaining a recurrent fast-weight state, but it consistently lags on in-context retrieval and long-context tasks.
arXiv:2609.13141v1 Announce Type: new Abstract: Post-training attention sparsification reduces the quadratic cumulative attention cost of pretrained Transformers by selecting a small set of context u...
arXiv:2607. 13395v1 Announce Type: new Abstract: The pursuit of autonomously self-improving models has attracted growing interest in the era of large-scale foundation models.
Attention-Aware Routing (AAR) augments the router in Mixture-of-Experts language models with temporal and spectral features derived from a sliding window of attention weights, thereby separating contextual information from the token’s hidden state. By keeping the base transformer frozen and training only routing parameters, AAR achieves a +3.37‑point improvement on GSM8K over a routing‑only baseline and demonstrates that routing changes propagate through the residual stream to reshape attention without directly updating the attention mechanism. The method also reduces long diverging generations, shows depth‑sensitivity affecting retrieval versus reasoning, and offers a controlled probe of routing‑relevant information across layers.
arXiv:2606. 18587v1 Announce Type: cross Abstract: Decoder-only Transformers compute attention over the KV cache of preceding tokens.
arXiv:2607. 19358v1 Announce Type: new Abstract: Recent advances in long chain-of-thought reasoning models such as DeepSeek-R1 have led to increasingly longer inference context lengths under the test-time scaling paradigm.
QueryFormer is a unified architecture designed for post‑click conversion rate prediction, addressing both feature interactions and sequential user behaviors. It introduces a stackable field–sequence block that generates query tokens via cross‑attention and packs sequence queries into shared‑parameter attention, improving efficiency and accuracy. The model won first place in the KDD Cup 2026 Tencent UniRec Challenge Industrial Track with an AUC of 0.83254, and scaling studies show that increasing view width slightly boosts validation AUC while maintaining low latency.
arXiv:2602. 03681v2 Announce Type: replace-cross Abstract: The quadratic computational complexity of softmax transformers has become a bottleneck in long-context scenarios.
The paper investigates how attention dynamics evolve across recurrent depth in language models, finding that attention support stabilizes early while hidden states and outputs take longer. It proposes WISE, a training‑free method that uses full attention in early steps and then reuses the discovered sparse working set for later steps, preserving performance on multi‑hop QA tasks. Experiments show that WISE maintains quality up to 2K context, offers measurable speedups, and highlights the importance of recurrent discovery of attention support.
arXiv:2406. 02021v3 Announce Type: replace-cross Abstract: Transformer, composed of self-attention and Feed-Forward Network, has revolutionized the landscape of network design across various vision tasks.
arXiv:2602.04852v3 Announce Type: replace Abstract: Linear attention offers a computationally efficient yet expressive alternative to softmax attention. However, recent empirical results indicate tha...